A World Bank course on AI-assisted coding with Positron, GitHub Copilot, and Stata.
Author: Eduard Bukin (ebukin@worldbank.org)
The purpose of this course is to equip economists who regularly use Stata and other modern statistical programming languages to leverage AI-assisted coding tools effectively and responsibly. The course introduces participants to AI coding assistants (GitHub Copilot) within the Positron IDE. The course consists of two 3-hour sessions with a self-study period in between.
By the end of the course, participants will be able to:
- Use AI assistants for code understanding, refactoring, and revision.
- Understand how LLMs work and what other concepts mean: tokens, prompts, context windows, agents
- Identify what AI can and cannot do reliably in a coding workflow
- Secure sensitive data and apply responsible AI guardrails
- Plan and execute complex, multi-step analytical tasks with an AI agent
- Apply context engineering techniques:
#tools,skills,/prompts,/agents, andMCPconnectors - Supervise AI agent execution and course-correct when it goes off track
Before the first session, participants complete:
- Software Setup — Stata 19+ MP, R 4.5+, Python 3.13+ with
uv, Positron, Quarto 1.9+, Git - GitHub & Copilot — personal GitHub account, WB org membership, Copilot access and premium requests
- Positron Extensions — Positron Stata (or Stata MCP), stataglow, databot
- Positron Assistant — enable assistant, connect to GitHub Copilot provider
| Time | Topic |
|---|---|
| 10 min | Welcome, introductions, course and materials overview |
| 30 min | Software overview: Positron, GitHub Copilot, and Stata setup |
| 40 min | AI in action with Stata (R): my typical data workflow using AI |
| 10 min | Break |
| 30 min | AI overview: how GitHub Copilot and LLMs work, key concepts, and capabilities |
| 30 min | Cookbook: securing sensitive data, guardrails, and responsible AI use |
| 30 min | Self-study exercises overview and Q&A |
Independent practice with structured case studies: reproduce old code, reproduce from an example, or try a new language.
| Time | Topic |
|---|---|
| 20 min | Q&A from self-study: challenges and discoveries |
| 30 min | Exercise: planning and executing complex tasks with parallel agentic workflows |
| 30 min | Context engineering: #tools, skills, MCP, and other relevant features |
| 10 min | Break |
| 45 min | Exercise: using the right #tools, integrating AI skills, developing /prompts and /agents |
| 30 min | Q&A and discussion |
| 5 min | Feedback survey |
| 10 min | Closing remarks and next steps |
This project is licensed under the MIT License together with the World Bank IGO Rider. The Rider is purely procedural: it reserves all privileges and immunities enjoyed by the World Bank, without adding restrictions to the MIT permissions. Please review both files before using, distributing or contributing.